Comparative study of induction motor fault analysis using feature extraction

摘要

Early fault detection in induction machines is essential to reduce downtime cost caused by unscheduled shut down of motor. It is also essential to protect the motor and to increase the lifetime machine component. In this work, a novel method has been designed to predict and classify the fault. Three phase stator currents collected from different known types of faulty motors and one healthy motor running as prime mover of a dc generator in load. The three phase stator currents have been used as raw data (amplitude vs time) for the analysis. By adding random bias signal the data samples (amplitude vs time) of six types unknown fault are also generated. The stator currents have been transformed by fast Fourier transform (FFT) method. The principal components have been computed on raw data as well as FFT spectrum. The relative distances have been calculated from scatter plot (PC-1 vs PC-2) of different faulty machines and the scatter plot of unknown faulty machines. Comparing the distance matrices, the unknown fault has been authenticated following minimum distance nearest neighborhood criterion. Comparing two types of scatter plots, one by PCA of raw data and the other by PCA of FFT spectrum, it has been concluded that sensitivity of FFT spectrum PCA yields is better.